Student adaptive auxiliary learning method and system based on artificial intelligence

Abstract

The invention provides a student adaptive auxiliary learning method and system based on artificial intelligence, and the method comprises the steps: constructing a subject knowledge graph, and carrying out the correlation and structuralization of knowledge points; the knowledge points are associated with learning resources and test questions in the subject knowledge graph; collecting learning behavior data of students, and constructing student portraits; the student portrait comprises three dimensions of learning style, knowledge level and hobbies and interests; wherein the knowledge level is a mastering probability of each knowledge point acquired according to the subject knowledge graph; personalized learning paths, learning resources and learning strategies are recommended to the students according to the student portraits and the subject knowledge maps; and learning results of the students are automatically evaluated and fed back. According to the characteristics and requirements of each student, a personalized learning scheme is provided, the learning efficiency is improved, the limitation of time and space is broken through, and the students can obtain high-quality learning resources which are more personalized for themselves anytime and anywhere.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence and educational informatization, and in particular relates to a student adaptive auxiliary learning method and system based on artificial intelligence. Background Art

[0002] Education is the foundation of a century-long plan. It is a means of disseminating knowledge and skills, carrying the responsibility and mission of nurturing the next generation. However, traditional education models are increasingly unable to meet the needs of the times.

[0003] The traditional education model is based on standardized teaching and generally adopts a "one-size-fits-all" teaching approach. It cannot meet students' individual learning needs, resulting in low learning efficiency, lack of timely and effective feedback, and low utilization of learning resources. This is specifically reflected in the following aspects:

[0004] 1. Unified progress and content cannot adapt to the differences in students' cognitive levels and learning speeds, resulting in "those with weak foundations unable to keep up, and those with outstanding abilities being restricted." The evaluation system relies too much on test scores, ignores multiple ability dimensions such as thinking ability and practical skills, and teaching feedback is seriously lagging behind.

[0005] 2. Relying on periodic examinations to discover learning problems leads to the long-term accumulation of knowledge gaps, high error correction costs, difficulty for teachers to track changes in each student's cognitive status in real time, and insufficient resource adaptability.

[0006] 3. Textbooks and exercise resources are rigid and cannot dynamically match students' current cognitive needs and interest preferences. There is a lack of integration of interdisciplinary knowledge, making it difficult for students to establish a systematic knowledge framework, and there is a lack of prior knowledge assessment.

[0007] 4. Failure to accurately diagnose the degree of prior knowledge acquisition results in the ineffectiveness of new knowledge learning due to weak foundation, forming a cognitive gap. Summary of the Invention

[0008] The present invention proposes an artificial intelligence-based student adaptive assisted learning method and system, which uses artificial intelligence technology to provide personalized assisted learning solutions according to the characteristics and needs of each student, thereby improving learning efficiency.

[0009] To achieve the above object, the technical solution of the present invention is achieved as follows:

[0010] An artificial intelligence-based student adaptive assisted learning method, comprising:

[0011] S1. Construct a subject knowledge graph to associate and structure knowledge points; knowledge points are associated with learning resources and test questions in the subject knowledge graph;

[0012] S2. Collect students' learning behavior data and construct student profiles; the student profiles include three dimensions: learning style, knowledge level, and interests and hobbies; the knowledge level is the mastery probability of each knowledge point obtained based on the subject knowledge graph;

[0013] S3. Recommend personalized learning paths, learning resources, and learning strategies to students based on their profiles and subject knowledge graphs.

[0014] S4. Automatically evaluate and provide feedback on students’ learning outcomes.

[0015] Furthermore, step S1 specifically includes:

[0016] S101. Determine the subject scope and the granularity of knowledge points of the subject knowledge graph, and define the entities, relationships, and attributes of the subject knowledge graph based on the subject scope and the granularity of knowledge points.

[0017] S102: Collect data sources, extract data, and complete the filling of the subject knowledge graph;

[0018] S103: Perform entity alignment, relationship conflict resolution, attribute fusion, and knowledge disambiguation to obtain a purified subject knowledge graph.

[0019] S104. The purified subject knowledge graph is mixedly stored, with the graph database storing entities and relationships, and the relational database storing attributes.

[0020] Furthermore, step S2 specifically includes:

[0021] S201. Collect students’ learning behavior data based on their authorization;

[0022] S202. Obtain students’ learning styles based on cluster analysis and learning activity sequence analysis;

[0023] S203. Based on the subject knowledge graph, the mastery probability of each knowledge point is obtained through the item response theory (IRT) model to obtain the student's knowledge level;

[0024] S204, extracting the student's interest topic vector through the natural language processing (NLP) model to obtain the student's interests and hobbies;

[0025] S205. The data of the three dimensions of students’ learning style, knowledge level, and interests and hobbies are integrated through a deep feature fusion network to generate a student portrait.

[0026] Furthermore, step S3 specifically includes:

[0027] S301: Build a recommendation engine, input the subject knowledge graph and the student profile of each student, and obtain current real-time learning information;

[0028] S302: The recommendation engine maps the knowledge level in the student profile to a subgraph in the subject knowledge graph. Based on the current real-time learning information, the recommendation engine runs a path search algorithm on the subject knowledge graph and performs personalized filtering to obtain recommended learning paths.

[0029] S303: The recommendation engine performs collaborative filtering of learning resources in the subject knowledge graph based on the learning style data, selects highly rated resources, extracts learning resource keywords through the NLP model, matches them with the interest data, and obtains recommendations for learning resources.

[0030] S304. The recommendation engine converts each level of the Bloom taxonomy into quantifiable judgment conditions, designs association rules, and obtains recommendations for learning strategies based on the association rules.

[0031] Furthermore, step S4 specifically includes:

[0032] S401, collect students’ learning achievement data and perform pre-processing;

[0033] S402: extracting knowledge points from the pre-processed learning outcome data based on entity linking and performing relationship matching analysis on the subject knowledge graph;

[0034] S403. Update the student portrait based on the analysis results, feed back the incorrect knowledge point matching relationship to the subject knowledge graph, and recommend new personalized solutions for students based on the updated student portrait and subject knowledge graph.

[0035] On the other hand, the present invention also proposes an artificial intelligence-based student adaptive auxiliary learning system, comprising:

[0036] Knowledge graph module: Builds a subject knowledge graph to associate and structure knowledge points; knowledge points are associated with learning resources and test questions in the subject knowledge graph;

[0037] Student portrait module: collects students' learning behavior data and constructs student portraits; the student portraits include three dimensions: learning style, knowledge level, and interests and hobbies; the knowledge level is the mastery probability of each knowledge point obtained from the subject knowledge graph;

[0038] Personalized recommendation module: recommends personalized learning paths, learning resources, and learning strategies to students based on their profiles and subject knowledge graphs;

[0039] Evaluation and feedback module: automatically evaluates and provides feedback on students' learning outcomes.

[0040] Furthermore, the knowledge graph module includes:

[0041] Definition unit: determines the subject scope and granularity of knowledge points of the subject knowledge graph, and defines the entities, relationships, and attributes of the subject knowledge graph based on the subject scope and granularity of knowledge points;

[0042] Data filling unit: collect data sources, extract data, and complete the filling of subject knowledge graph;

[0043] Purification unit: performs entity alignment, relationship conflict resolution, attribute fusion, and knowledge disambiguation to obtain a purified subject knowledge graph;

[0044] Storage unit: Hybrid storage of purified subject knowledge graphs, with graph databases storing entities and relationships, and relational databases storing attributes.

[0045] Furthermore, the student portrait module includes:

[0046] Authorization collection unit: collects students' learning behavior data based on their authorization;

[0047] Style analysis unit: Based on cluster analysis and learning activity sequence analysis, obtain students' learning styles;

[0048] Knowledge level unit: Based on the subject knowledge map, the mastery probability of each knowledge point is obtained through the item response theory (IRT) model to obtain the student's knowledge level;

[0049] Interest and hobbies unit: Through the natural language processing (NLP) model, the student's interest theme vector is extracted to obtain the student's interests and hobbies;

[0050] Portrait unit: The data of three dimensions of students' learning style, knowledge level, and interests and hobbies are integrated through a deep feature fusion network to generate student portraits.

[0051] Furthermore, the personalized recommendation module includes:

[0052] Input unit: Build a recommendation engine, input subject knowledge graphs and student profiles, and obtain current real-time learning information;

[0053] Path recommendation unit: The recommendation engine maps the knowledge level in the student profile to a subgraph in the subject knowledge graph. Based on the current real-time learning information, it runs a path search algorithm on the subject knowledge graph and performs personalized filtering to obtain recommended learning paths.

[0054] Resource recommendation unit: The recommendation engine performs collaborative filtering of learning resources in the subject knowledge graph based on learning style data, selects highly rated resources, extracts learning resource keywords through the NLP model, and matches them with interest data to obtain learning resource recommendations;

[0055] Strategy recommendation unit: The recommendation engine converts each level of Bloom's taxonomy into quantifiable judgment conditions, designs association rules, and recommends learning strategies based on the association rules.

[0056] Furthermore, the evaluation feedback module includes:

[0057] Achievement collection unit: collect students' learning achievement data and pre-process them;

[0058] Matching analysis unit: extracts knowledge points from the pre-processed learning outcome data based on entity links and performs relationship matching analysis with the subject knowledge graph;

[0059] Feedback unit: Updates the student portrait based on the analysis results, feeds back incorrect knowledge point matching relationships to the subject knowledge graph, and recommends new personalized plans for students based on the updated student portrait and subject knowledge graph.

[0060] Compared with the existing technology, the student adaptive assisted learning method and system based on artificial intelligence proposed in the present invention has the following beneficial effects:

[0061] 1. The present invention provides personalized learning plans based on the characteristics and needs of each student, improves learning efficiency, and breaks the limitations of time and space, allowing students to obtain high-quality learning resources that are more personalized to themselves anytime and anywhere; it uses artificial intelligence technology to automatically evaluate students' learning outcomes and provide detailed feedback to help students identify and fill in the gaps.

[0062] 2. The subject knowledge graph solution proposed in this invention organizes scattered subject knowledge points in a systematic, hierarchical and associative manner, and clearly defines various semantic relationships between knowledge points (such as belonging to, prerequisite, dependency, similarity, application, etc.), providing a structured knowledge foundation for knowledge level assessment (IRT / BKT / DKT), personalized path planning, resource labeling and recommendation, and learning situation diagnosis.

[0063] 3. The student portrait solution proposed in this invention systematically collects, processes, and analyzes multi-dimensional learning behavior data to construct a dynamic, multi-dimensional, and interpretable student personal portrait, providing an accurate student cognition and preference model for intelligent teaching.

[0064] 4. The personalized recommendation scheme proposed by the present invention recommends the most suitable content according to students' knowledge level, learning style, and interests, and updates the recommendation strategy in real time as the learning progresses, while achieving a balance between goals such as knowledge coverage, ability improvement, and interest stimulation.

[0065] 5. The present invention uses artificial intelligence technology to automatically evaluate students' learning outcomes, forming a complete closed loop with student portraits, knowledge graphs and adaptive recommendation systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 This is a schematic diagram of the overall process of Example 1 of the present invention;

[0067] Figure 2 Schematic diagram of the personalized recommendation process of Example 1 of the present invention;

[0068] Figure 3 This is a schematic diagram of learning achievement evaluation feedback in Example 1 of the present invention. DETAILED DESCRIPTION

[0069] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0070] In order to make the purpose and features of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are further described below with reference to the accompanying drawings.

[0071] Example 1:

[0072] The student adaptive auxiliary learning method based on artificial intelligence implemented in this embodiment is as follows: Figure 1 Shown, including:

[0073] 1. Build a subject knowledge map to associate and structure knowledge points; knowledge points are associated with learning resources and test questions in the subject knowledge map.

[0074] This step is mainly divided into four stages.

[0075] Phase (1): Knowledge modeling, defining the skeleton of the graph.

[0076] 1. Determine the scope and granularity:

[0077] Clarify the subject scope of the map (e.g., junior high school math, high school physics, elementary school Chinese). Determine the granularity of the knowledge points: typically, use the "sections" or core concepts in the curriculum standards or textbook catalog as the basic units (e.g., "quadratic equations," "Newton's First Law," "metaphors"). Avoid being too macro (e.g., "algebra") or too micro (e.g., "the specific steps for moving a term").

[0078] 2. Constructing Ontology - defining knowledge "types" and "relationships":

[0079] (1) Core concepts include:

[0080] Knowledge point (Concept): core entity, representing the basic concepts, principles, and skills in a subject.

[0081] Domain / Module: The upper-level organizational unit of knowledge points (such as "Number and Algebra", "Mechanics", "Modern Chinese Reading").

[0082] Ability dimension: such as "understanding", "application", "analysis", and "creation" (refer to Bloom's taxonomy).

[0083] Resource: Learning resources (videos, documents, exercises, experiments) associated with knowledge points.

[0084] Question: A question used to evaluate knowledge points.

[0085] Teaching Standards: Related to national / local curriculum standards.

[0086] (2) Core relationships include:

[0087] Belongs to (isPartOf): Knowledge point A belongs to domain / module B (concept -> domain).

[0088] Prerequisite: Mastering knowledge point A is a prerequisite for learning knowledge point B (Concept A -> Concept B).

[0089] Dependency (dependsOn): In-depth understanding of knowledge point B requires the support of knowledge point A (weaker than the prerequisite, may be learned in parallel but with dependency) (Concept A -> Concept B).

[0090] Similarity (similarTo): Knowledge points A and B are highly related in concepts or methods (Concept A <-> Concept B).

[0091] Application (appliesTo): Knowledge point A is a specific application of knowledge point B (Concept A -> Concept B) or knowledge point A can be applied to field C (Concept A -> Field C).

[0092] Assessment (assessedBy): Knowledge point A can be assessed by question D (Concept A -> Question D).

[0093] TaughtBy: Knowledge point A can be learned through resource E (Concept A->Resource E).

[0094] CorrespondsTo: Knowledge point A corresponds to curriculum standard item F (Concept A->Standard F).

[0095] Requires Ability: Mastering knowledge point A requires reaching ability dimension G (concept A->ability G).

[0096] (3) Attributes:

[0097] Knowledge point: Attributes include unique identifier (id), name (name), description (description), difficulty (difficulty), importance (importance), and creation / update time.

[0098] Test question: Attributes include id, content, answer, analysis, difficulty, discrimination, and knowledge point label (multi-label).

[0099] Resource: Attributes include id, title, type, URL / path, description, applicable object, and tags.

[0100] (4) Tools:

[0101] In this embodiment, ontology modeling tools such as Protégé tool software are used to construct ontology concept classes, relationships, attributes and instances.

[0102] Phase (2): Knowledge acquisition and extraction, filling the graph with flesh and blood.

[0103] 1. Data source:

[0104] The data for knowledge acquisition and extraction can be structured / semi-structured data, and the data sources come from: national / local curriculum standard documents (extracting knowledge modules and ability requirements); authoritative textbook catalogs and chapter structures (extracting knowledge point hierarchical relationships); knowledge trees and mind maps of teaching materials; knowledge point tags in existing question bank systems (need to be cleaned and mapped); metadata of learning resource libraries (knowledge point tags), etc.

[0105] Unstructured data can also be used, which comes from: textbook content, teacher lesson plans, teaching instructions, high-quality academic papers, encyclopedia entries (which can be used to enrich descriptions and relationships), course outlines and descriptions of online education platforms, etc.

[0106] 2. Knowledge extraction:

[0107] For structured / semi-structured data, based on specific formats (such as textbook catalogs), rules are written to parse the tree structure, extract the belonging relationships, parse the knowledge tables in the textbooks or standards, and import the existing knowledge point label data in the question bank and resource library for cleaning and mapping.

[0108] For unstructured text data, we identify the names of knowledge points (concept entities) mentioned in the text, then perform relationship extraction. We define language patterns to identify specific relationships (e.g., "To learn A, you need to master B first" -> B must first learn A; "A is the foundation of B" -> A must first learn B; "A and B are similar" -> A is similar to B). We can also extract knowledge by training models (e.g., BERT, RoBERTa+relation classification layer) with large amounts of annotated corpus, or automatically generate training data using knowledge bases or structured data to train models to predict new relationships.

[0109] Phase (3): Knowledge fusion and disambiguation to ensure the purity of the graph.

[0110] Entity alignment: Identify different names or descriptions of the same knowledge point in different sources (e.g., "Pythagorean Theorem" vs. "Pythagorean Theorem") based on name similarity (edit distance, Jaccard), attribute similarity (cosine similarity of description text embedding vectors), contextual similarity, and rules (synonym tables). Entity alignment can be accomplished using Dedupe (a Python library) or an entity alignment model based on embeddings.

[0111] Relationship conflict resolution: To resolve the contradictions in the descriptions of the relationship between the same pair of knowledge points from different sources (e.g., source A says that A takes B first, while source B says that B takes A first), you can set the priority of the data source (e.g., curriculum standards > authoritative textbooks > teaching aids > automatic extraction) and use the data source with the highest priority to describe the relationship between knowledge points, or conduct expert arbitration.

[0112] Attribute fusion: Merge attributes from different sources of the same entity to resolve conflicts (take the one with higher priority, take the latest, or mark multiple possibilities).

[0113] Knowledge disambiguation: Dealing with homonyms (e.g., the different meanings of "vector" in mathematics and physics). This is usually resolved by clarifying the context (the field to which it belongs) or adding qualifiers to the entity name (e.g., "vector (mathematics)", "vector (physics)").

[0114] After the above processing, a unified and unambiguous knowledge point entity library and a consistent relationship network can be output.

[0115] Phase (IV): Knowledge storage - building a graph warehouse.

[0116] Graph databases are specifically designed for storing and querying relational networks and efficiently support graph traversal queries. Therefore, in this embodiment, core relationships and graph queries are stored through the graph database. Relational databases are suitable for small-scale or simple-relationship graphs. This embodiment is used to store detailed attribute information to form a hybrid storage solution.

[0117] 2. Collect students' learning behavior data and construct student portraits; the student portraits include three dimensions: learning style, knowledge level, and interests and hobbies; the knowledge level is the probability of mastering each knowledge point obtained based on the subject knowledge graph.

[0118] This step specifically includes:

[0119] 1. Collect students’ learning behavior data based on their authorization;

[0120] On the basis of obtaining students' authorization, the operation data of students using the school's online learning platform can be collected, such as data related to learning style, including resource preferences: access resource type (video, text, PPT, audio, interactive simulation), number of views, duration (average viewing time, whether to fast forward / replay / pause), and download behavior; interaction mode: number of attempts in interactive exercises, quizzes, and simulations, operation sequence (do you watch the explanation first or try it first?), preference for independent completion or seeking help (frequency of clicking prompts); time mode: active learning time period (morning, afternoon, evening, late at night), single learning duration, regularity of learning intervals; social learning: frequency of participation in discussion forums (posting, replying, liking), number of active questions, frequency and quality of answering others' questions (text analysis basis); note-taking behavior: whether to take notes, note location (highlighting, page margin notes), note content characteristics (keyword extraction, summary vs. detail); knowledge level related data can also be collected, including assessment performance: scores of all tests / exams, answering time (overall and per question), Number of attempts (especially adaptive exercises), record of wrong questions (specific questions, options, knowledge point annotations); learning progress: completion progress of courses / chapter / knowledge points, completion speed (compared with class average or preset standards), number of times specific knowledge points are repeated; practice behavior: accuracy rate of practice questions (classified by knowledge points), practice time, frequency of revealing hints or answers, and whether to actively challenge more difficult questions; knowledge application: depth, breadth, and innovation of knowledge application demonstrated in projects, homework, and case analysis (text / code analysis); data related to interests and hobbies can also be collected, including independent exploration: actively searched learning resource keywords, browsed non-compulsory courses / content, and favorites content analysis; extended learning: completion and involvement in extracurricular challenge questions, extended reading materials, and recommended reading; project / activity selection: tendency to choose among optional projects / topics (such as choosing AI direction for programming projects and science fiction themes for writing); community participation: most frequently participated topic sections in the discussion area, types of other users (experts / classmates) followed, and created topic themes (text topic modeling).

[0121] In addition to the above data, relevant data from intelligent teaching applications / tools (such as question bank apps, programming environments, and virtual laboratories) can also be collected, including learning style-related data: interaction preferences within the tool (graphical operations vs. command lines), problem-solving paths (trial-and-error vs. planning-and-execution), and ways of seeking help (consulting documents vs. asking questions directly); knowledge level-related data: code debugging behavior (error type, repair speed), number of iterations of experimental design modifications, problem-solving efficiency, and strategy complexity; and hobbies-related data: non-prescribed functions explored within the tool, personalized project themes created, and thematic tendencies of shared works.

[0122] In addition, self-reported student data can also be collected: for example, through learning style questionnaires, collected regularly or initially using validated scales (such as the Felder-Silverman Index of Learning Styles); or through interest surveys, such as open-ended questions or the selection of interest labels (subjects, activities, future directions, etc.).

[0123] The above collection process can use tracking technology to embed code at key interaction points (such as page views, button clicks, video operations, exercise submissions, and discussion posts) within platforms, applications, and tools to record user IDs, timestamps, event types, and event details (such as question IDs, options, resource IDs, and operation types). Alternatively, APIs can be used to obtain structured data (such as grades, progress, and discussion post content).

[0124] The aforementioned mobile phone process is premised on ensuring data privacy and security. First, data collection must be minimized, collecting only the necessary data directly relevant to the profiling objective. Second, clear notification and authorization must be ensured. Students must be clearly informed of the purpose, scope, and intended use of the data, and explicit authorization must be obtained (especially for sensitive or in-depth analysis). Access control and encryption must be implemented for the collected data.

[0125] The collected data must be preprocessed, including data cleaning: processing missing values ​​(interpolation or marking), outliers (detection, analysis, correction or elimination), duplicate values, format errors, etc., and then associating data from different sources to the same student ID, and then performing timestamp processing (extracting time periods and durations), categorical variable encoding, and text segmentation (discussion posts, notes, homework).

[0126] 2. Obtain students’ learning styles based on cluster analysis and learning activity sequence analysis;

[0127] After data preprocessing, cluster analysis is performed based on behavioral characteristics (resource preferences, interaction patterns, time patterns, and social note-taking) to group students into "style clusters." Examples include "visually dominant - independent exploration," "text-preferred - social discussion," and "fragmented learning - practice-oriented." Learning activity sequences are then analyzed (e.g., [watch video] -> [do exercises] -> [check mistakes] vs. [try exercises] -> [watch explanations of mistakes] -> [redo]) to identify typical learning path patterns. Based on predefined or learned rules, these behavior combinations are mapped to style labels (e.g., IF video viewing time > 70% AND help-seeking frequency < 0.1 THEN style = visually dominant - independent). Model calibration is then performed using the questionnaire: Using the questionnaire results as a supervisory signal, a classification model (e.g., SVM, random forest, or simple neural network) is trained to predict style labels for students who did not complete the questionnaire, or to calibrate the clustering results. Based on the clustering results, each student is assigned to their respective learning style, generating feature vectors / values ​​for the learning style dimensions.

[0128] 3. Based on the subject knowledge graph, the mastery probability of each knowledge point is obtained through the item response theory (IRT) model to obtain the student's knowledge level;

[0129] Based on the subject knowledge graph constructed in step 1 above, clarify the hierarchical relationships and dependencies between knowledge points. Based on item response theory (IRT), using models such as the Rasch model and the 2PL / 3PL model, estimate the student's ability value θ and the difficulty b, discrimination a, and guessability c of the knowledge point based on their responses, providing the student's probability of mastering the knowledge point. Deep learning models such as RNN / LSTM / Transformer can also be used to predict future performance and the probability of mastering each knowledge point based on historical answer sequences, capturing more complex dependencies. Analyze incorrect answers, combining the knowledge graph and the error type library to diagnose the specific causes of errors (e.g., conceptual confusion, calculation errors, misunderstandings, carelessness). Finally, through time series analysis, analyze the changing trends in knowledge mastery (progress, stagnation, regression), ultimately obtaining the characteristic vector / value of the student's knowledge level dimension.

[0130] 4. Extract students’ interest topic vectors and obtain their interests and hobbies through the natural language processing (NLP) model;

[0131] Through the natural language processing (NLP) model, implicit topic distribution is extracted from discussion posts, notes, homework, and search terms to form an "interest topic vector", and then keyword extraction is performed to identify high-frequency and key interest words.

[0132] Based on the keywords of the topic, sentiment analysis is performed to analyze the emotional tendency (positive, negative) towards specific topics / content. Collaborative filtering (CF) is then performed, including user-based filtering, for example: "Students with similar behavior to you are also interested in X"; item-based filtering, for example: "Students who like resource A usually also like resource B." Then, the following steps are performed in sequence: content analysis: analyzing the metadata (topic tags, classification, difficulty, author) of the resources that students actively access, collect, and have high engagement with; network analysis: analyzing the relationships and interactive objects of students in the learning community to identify interest communities; association rule mining (Apriori, FP-Growth): discovering interest combinations (such as "Students who like AI also often read Frontiers of Mathematics"). After the above analysis process, the feature vectors / values ​​of the student's interest and hobby dimensions can be obtained.

[0133] 5. The data of three dimensions, namely, students’ learning style, knowledge level, and interests and hobbies, are integrated through a deep feature fusion network to generate student portraits.

[0134] After obtaining feature vectors / values ​​for the three dimensions of a student's learning style, knowledge level, and interests and hobbies, advanced models such as GNNs, multi-task learning models, and deep feature fusion networks are used to explore the inherent connections between features across these dimensions, generating more unified and richer student profile data. The resulting profile data is stored in a database using a defined JSON Schema (or similar structure) for query and update. Confidence and coverage calculations can be used to assess the credibility of the profile results and the extent of their dimensional coverage.

[0135] 3. Recommend personalized learning paths, learning resources and learning strategies to students based on student portraits and subject knowledge maps.

[0136] The specific method of this step is as follows Figure 2 Shown, including:

[0137] 1. Build a recommendation engine, input subject knowledge graphs and student profiles, and obtain current real-time learning information;

[0138] The recommendation engine is built using the distributed streaming processing framework Flink technology. The recommendation engine receives information input such as subject knowledge graphs, student portraits, and current real-time learning information, analyzes the input information, and provides recommendations on learning paths, learning resources, and learning strategies.

[0139] Input a student profile and use it to obtain the student's knowledge level: probability of mastering knowledge points, list of weaknesses, and ability trends; learning style: resource preference (video / text), interaction mode (independent / collaborative), and time mode; interests and hobbies: academic interest topics and activity preferences.

[0140] Input the subject knowledge graph and obtain knowledge point relationships (prerequisites, dependencies, similarities), resource associations (knowledge point labels marked in videos, exercises, experiments, etc.), and knowledge point metadata (difficulty, importance, and ability dimensions).

[0141] By collecting students' current learning actions (such as starting to learn a certain knowledge point K through the online learning platform, submitting a certain homework Q, etc.), as well as recent answer records (correctness rate, time taken, etc.), we can obtain students' current real-time learning information.

[0142] 2. The recommendation engine maps the knowledge level in the student profile to a subgraph in the subject knowledge graph. Based on the current real-time learning information, it runs a path search algorithm on the subject knowledge graph and performs personalized filtering to obtain recommended learning paths.

[0143] The path search algorithm can be based on the prerequisite relationship and use the improved Dijkstra algorithm (weight = difficulty of knowledge point × current level gap of students).

[0144] Personalized filtering refers to personalized selection based on the student's learning style, knowledge level, and interests. For example, if the student is visually inclined, paths containing video resources will be prioritized. If the student's interest is AI, relevant expanded knowledge points (discovered through similar relationships) will be inserted into the path.

[0145] 3. The recommendation engine performs collaborative filtering of learning resources in the subject knowledge graph based on learning style data, selects highly rated resources, extracts learning resource keywords through the NLP model, and matches them with interest data to obtain recommendations for learning resources.

[0146] The collaborative filtering can be understood as follows: for example, if the student profiles of students A and B are similar, and student B has a highly rated resource R, then R can be recommended to student A.

[0147] 4. The recommendation engine converts each level of Bloom's taxonomy into quantifiable judgment conditions, designs association rules, and recommends learning strategies based on the association rules.

[0148] The six levels of the Loom taxonomy are clearly defined: memory, comprehension, application, analysis, evaluation, and creation. These levels represent varying degrees of learning depth and complexity. Quantifiable criteria are designed for each level. For example, at the memory level, this is quantified by the number of knowledge point repetitions and memory accuracy (e.g., correctness in fill-in-the-blank questions), with a threshold of 80% accuracy. At the comprehension level, this is measured by the completeness of concept recitation and the score on analogy questions, requiring the ability to explain 70% of the core elements. At the application level, this requires a 60% or higher solution rate for scenario simulation questions, or 3 or higher correct case transfers. At the analysis level, this requires a logical integrity score for element decomposition (e.g., a mind map structure score of 4 / 5). At the evaluation level, this requires 3 or more supporting arguments, with a 2.5 / 3 or higher rebuttal validity score. At the creation level, this requires a 50% or higher pass rate for original proposals.

[0149] Design association rules and get recommendations for learning strategies based on the association rules. For example, if the accuracy of the memory layer is ≥85% + the understanding layer score is <60%, visual aids can be provided to students.

[0150] 4. Automatically evaluate and provide feedback on students’ learning outcomes.

[0151] This step specifically includes:

[0152] 1. Collect students’ learning outcome data and pre-process them;

[0153] This can be collected through online learning platforms or other similar teaching tools, and the collected content includes:

[0154] Structured answer data (based on online question banks / exam systems); text answers (derived from essays / short answers / programming notes); voice answers (derived from oral practice recordings, etc.); screen recordings of the problem-solving process (achieved through screen capture); collaborative discussion records (derived from online discussion forum logs, etc.).

[0155] The collected data is preprocessed. In addition to missing value filling and outlier detection, text data is subject to word segmentation, stop word removal, and syntactic parsing; voice data is subject to speech-to-text (ASR) and voiceprint segmentation; key frame extraction and operation sequence encoding are performed on screen recording data; and dialogue role separation and sentiment polarity labeling are performed on discussion records.

[0156] 2. Extract knowledge points from the pre-processed learning outcome data based on entity links and perform relationship matching analysis with the subject knowledge graph;

[0157] The entity linking uses a joint model of BERT+knowledge graph embedding (such as BLINK), which can identify candidate knowledge point entities in the preprocessed learning outcome data; the relationship matching analysis obtains the knowledge point entities in the subject knowledge graph based on name similarity, calculates the contextual semantic similarity, and builds a rule engine based on the prerequisite and dependency relationships of the subject knowledge graph, and uses Cypher query statements to mine potential relationship paths for matching analysis.

[0158] 3. Update the student portrait based on the analysis results, feedback the incorrect knowledge point matching relationship to the subject knowledge graph, and recommend new personalized plans for students based on the updated student portrait and subject knowledge graph.

[0159] like Figure 3 As shown, the student's learning outcomes are evaluated and analyzed by the evaluation system used to implement this step, that is, the learning outcome data of step 2 is matched with the knowledge points of the subject knowledge graph, and the characteristic vector / value of the student's new knowledge level dimension is obtained and sent to the system for implementing student portraits, and the vector of the student's knowledge mastery is updated, thereby updating the student portrait; the erroneous relationship obtained by the analysis is fed back to the subject knowledge graph, for example, if 50% of students confuse concepts A and B, the relationship between knowledge points A and B is updated in the subject knowledge graph; the update of student portraits and subject knowledge graphs triggers new resource matching, and the recommendation engine used for personalized recommendations recommends new personalized plans to students to remedy deficiencies in learning.

[0160] The adaptive assisted learning method proposed in this embodiment provides personalized learning plans based on the characteristics and needs of each student, improves learning efficiency, and breaks the limitations of time and space, allowing students to obtain high-quality learning resources that are more personalized to themselves anytime and anywhere; it uses artificial intelligence technology to automatically evaluate students' learning outcomes and provide detailed feedback to help students identify and fill in the gaps.

[0161] Example 2:

[0162] This embodiment proposes an artificial intelligence-based student adaptive auxiliary learning system, including:

[0163] Knowledge graph module: Builds a subject knowledge graph to associate and structure knowledge points; knowledge points are associated with learning resources and test questions in the subject knowledge graph;

[0164] Student portrait module: collects students' learning behavior data and constructs student portraits; the student portraits include three dimensions: learning style, knowledge level, and interests and hobbies; the knowledge level is the mastery probability of each knowledge point obtained from the subject knowledge graph;

[0165] Personalized recommendation module: recommends personalized learning paths, learning resources, and learning strategies to students based on their profiles and subject knowledge graphs;

[0166] Evaluation and feedback module: automatically evaluates and provides feedback on students' learning outcomes.

[0167] Among them, the knowledge graph module includes:

[0168] Definition unit: determines the subject scope and granularity of knowledge points of the subject knowledge graph, and defines the entities, relationships, and attributes of the subject knowledge graph based on the subject scope and granularity of knowledge points;

[0169] Data filling unit: collect data sources, extract data, and complete the filling of subject knowledge graph;

[0170] Purification unit: performs entity alignment, relationship conflict resolution, attribute fusion, and knowledge disambiguation to obtain a purified subject knowledge graph;

[0171] Storage unit: Hybrid storage of purified subject knowledge graphs, with graph databases storing entities and relationships, and relational databases storing attributes.

[0172] The student portrait module includes:

[0173] Authorization collection unit: collects students' learning behavior data based on their authorization;

[0174] Style analysis unit: Based on cluster analysis and learning activity sequence analysis, obtain students' learning styles;

[0175] Knowledge level unit: Based on the subject knowledge map, the mastery probability of each knowledge point is obtained through the item response theory (IRT) model to obtain the student's knowledge level;

[0176] Interest and hobbies unit: Through the natural language processing (NLP) model, the student's interest theme vector is extracted to obtain the student's interests and hobbies;

[0177] Portrait unit: The data of three dimensions of students' learning style, knowledge level, and interests and hobbies are integrated through a deep feature fusion network to generate student portraits.

[0178] The personalized recommendation module includes:

[0179] Input unit: Build a recommendation engine, input subject knowledge graphs and student profiles, and obtain current real-time learning information;

[0180] Path recommendation unit: The recommendation engine maps the knowledge level in the student profile to a subgraph in the subject knowledge graph. Based on the current real-time learning information, it runs a path search algorithm on the subject knowledge graph and performs personalized filtering to obtain recommended learning paths.

[0181] Resource recommendation unit: The recommendation engine performs collaborative filtering of learning resources in the subject knowledge graph based on learning style data, selects highly rated resources, extracts learning resource keywords through the NLP model, and matches them with interest data to obtain learning resource recommendations;

[0182] Strategy recommendation unit: The recommendation engine converts each level of Bloom's taxonomy into quantifiable judgment conditions, designs association rules, and recommends learning strategies based on the association rules.

[0183] The assessment feedback module includes:

[0184] Achievement collection unit: collect students' learning achievement data and pre-process them;

[0185] Matching analysis unit: extracts knowledge points from the pre-processed learning outcome data based on entity links and performs relationship matching analysis with the subject knowledge graph;

[0186] Feedback unit: Updates the student portrait based on the analysis results, feeds back incorrect knowledge point matching relationships to the subject knowledge graph, and recommends new personalized plans for students based on the updated student portrait and subject knowledge graph.

[0187] The student adaptive assisted learning system based on artificial intelligence proposed in this embodiment can implement the student adaptive assisted learning method based on artificial intelligence proposed in Example 1, and can achieve the same beneficial effects as Example 1.

[0188] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A student-adaptive assisted learning method based on artificial intelligence, characterized in that: include: S1. Construct a subject knowledge graph to associate and structure knowledge points; Knowledge points are associated with learning resources and test questions in the subject knowledge graph; S2. Collect students' learning behavior data and construct student profiles; the student profiles include three dimensions: learning style, knowledge level, and interests and hobbies; the knowledge level is the mastery probability of each knowledge point obtained based on the subject knowledge graph; S3. Recommend personalized learning paths, learning resources, and learning strategies to students based on their profiles and subject knowledge graphs. S4. Automatically evaluate and provide feedback on students’ learning outcomes.

2. The student-adaptive assisted learning method based on artificial intelligence according to claim 1 is characterized in that: Step S1 specifically includes: S101. Determine the subject scope and the granularity of knowledge points of the subject knowledge graph, and define the entities, relationships, and attributes of the subject knowledge graph based on the subject scope and the granularity of knowledge points. S102: Collect data sources, extract data, and complete the filling of the subject knowledge graph; S103: Perform entity alignment, relationship conflict resolution, attribute fusion, and knowledge disambiguation to obtain a purified subject knowledge graph. S104. The purified subject knowledge graph is stored in a mixed manner, with the graph database storing entities and relationships, and the relational database storing attributes.

3. The student adaptive assisted learning method based on artificial intelligence according to claim 1 is characterized in that: Step S2 specifically includes: S201. Collect students’ learning behavior data based on their authorization; S202. Obtain students’ learning styles based on cluster analysis and learning activity sequence analysis; S203. Based on the subject knowledge graph, the mastery probability of each knowledge point is obtained through the item response theory (IRT) model to obtain the student's knowledge level; S204, extracting the student's interest topic vector through the natural language processing (NLP) model to obtain the student's interests and hobbies; S205. The data of the three dimensions of students’ learning style, knowledge level, and interests and hobbies are integrated through a deep feature fusion network to generate a student portrait.

4. The student-adaptive assisted learning method based on artificial intelligence according to claim 1 is characterized in that: Step S3 specifically includes: S301: Build a recommendation engine, input the subject knowledge graph and the student profile of each student, and obtain current real-time learning information; S302: The recommendation engine maps the knowledge level in the student profile to a subgraph in the subject knowledge graph. Based on the current real-time learning information, the recommendation engine runs a path search algorithm on the subject knowledge graph and performs personalized filtering to obtain recommended learning paths. S303: The recommendation engine performs collaborative filtering of learning resources in the subject knowledge graph based on the learning style data, selects highly rated resources, extracts learning resource keywords through the NLP model, matches them with the interest data, and obtains recommendations for learning resources. S304. The recommendation engine converts each level of the Bloom taxonomy into quantifiable judgment conditions, designs association rules, and obtains recommendations for learning strategies based on the association rules.

5. The student adaptive assisted learning method based on artificial intelligence according to claim 1 is characterized in that: Step S4 specifically includes: S401, collect students’ learning achievement data and perform pre-processing; S402: extracting knowledge points from the pre-processed learning outcome data based on entity linking and performing relationship matching analysis on the subject knowledge graph; S403. Update the student portrait based on the analysis results, feed back the incorrect knowledge point matching relationship to the subject knowledge graph, and recommend new personalized solutions for students based on the updated student portrait and subject knowledge graph.

6. An artificial intelligence-based student adaptive learning system, characterized in that: include: Knowledge graph module: builds subject knowledge graphs to associate and structure knowledge points; Knowledge points are associated with learning resources and test questions in the subject knowledge graph; Student portrait module: collects students' learning behavior data and constructs student portraits; the student portraits include three dimensions: learning style, knowledge level, and interests and hobbies; the knowledge level is the mastery probability of each knowledge point obtained from the subject knowledge graph; Personalized recommendation module: recommends personalized learning paths, learning resources, and learning strategies to students based on their profiles and subject knowledge graphs; Evaluation and feedback module: automatically evaluates and provides feedback on students' learning outcomes.

7. The student adaptive auxiliary learning system based on artificial intelligence according to claim 6 is characterized in that: The knowledge graph module includes: Definition unit: determines the subject scope and granularity of knowledge points of the subject knowledge graph, and defines the entities, relationships, and attributes of the subject knowledge graph based on the subject scope and granularity of knowledge points; Data filling unit: collect data sources, extract data, and complete the filling of subject knowledge graph; Purification unit: performs entity alignment, relationship conflict resolution, attribute fusion, and knowledge disambiguation to obtain a purified subject knowledge graph; Storage unit: Hybrid storage of purified subject knowledge graphs, with graph databases storing entities and relationships, and relational databases storing attributes.

8. The student adaptive auxiliary learning system based on artificial intelligence according to claim 6 is characterized in that: The student portrait module includes: Authorization collection unit: collects students' learning behavior data based on their authorization; Style analysis unit: Based on cluster analysis and learning activity sequence analysis, obtain students' learning styles; Knowledge level unit: Based on the subject knowledge map, the mastery probability of each knowledge point is obtained through the item response theory (IRT) model to obtain the student's knowledge level; Interest and hobbies unit: Through the natural language processing (NLP) model, the student's interest theme vector is extracted to obtain the student's interests and hobbies; Portrait unit: The data of three dimensions of students' learning style, knowledge level, and interests and hobbies are integrated through a deep feature fusion network to generate student portraits.

9. The student adaptive auxiliary learning system based on artificial intelligence according to claim 6 is characterized in that: The personalized recommendation module includes: Input unit: Build a recommendation engine, input subject knowledge graphs and student profiles, and obtain current real-time learning information; Path recommendation unit: The recommendation engine maps the knowledge level in the student profile to a subgraph in the subject knowledge graph. Based on the current real-time learning information, it runs a path search algorithm on the subject knowledge graph and performs personalized filtering to obtain recommended learning paths. Resource recommendation unit: The recommendation engine performs collaborative filtering of learning resources in the subject knowledge graph based on learning style data, selects highly rated resources, extracts learning resource keywords through the NLP model, and matches them with interest data to obtain learning resource recommendations; Strategy recommendation unit: The recommendation engine converts each level of Bloom's taxonomy into quantifiable judgment conditions, designs association rules, and recommends learning strategies based on the association rules.

10. The student adaptive auxiliary learning system based on artificial intelligence according to claim 6 is characterized in that: The assessment feedback module includes: Achievement collection unit: collect students' learning achievement data and pre-process them; Matching analysis unit: extracts knowledge points from the pre-processed learning outcome data based on entity links and performs relationship matching analysis with the subject knowledge graph; Feedback unit: Updates the student portrait based on the analysis results, feeds back incorrect knowledge point matching relationships to the subject knowledge graph, and recommends new personalized plans for students based on the updated student portrait and subject knowledge graph.

Images